04.02.2015 Views

Stochastic Programming - Index of

Stochastic Programming - Index of

Stochastic Programming - Index of

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

RECOURSE PROBLEMS 187<br />

means <strong>of</strong> the following problem, where we calculate what is left for the next<br />

random variable:<br />

α r+1<br />

i<br />

= α r i − min{yr+ i<br />

,y r−<br />

i<br />

, 0}. (4.6)<br />

What we are doing here is to find, for each variable, how much ˜ξ r ,intheworst<br />

case, uses <strong>of</strong> the bound on variable i in the negative direction. That is then<br />

subtracted <strong>of</strong>f what we had before. There are three possibilities. Both (4.4) and<br />

(4.5) may yield non-negative values for the variable y i . In that case nothing<br />

is used <strong>of</strong> the available “negative bound” α r i .Thenαr+1 i = α r i . Alternatively,<br />

if (4.4) has yi r+ < 0, then it will in the worst case use yi<br />

r+ <strong>of</strong> the available<br />

“negative bound”. Finally, if (4.5) has y r−<br />

i < 0thenintheworstcaseweuse<br />

y r−<br />

i <strong>of</strong> the bound. Therefore α r+1<br />

i is what is left for the next random variable.<br />

Similarly, we find<br />

β r+1<br />

i<br />

= βi r − max{yr+ i<br />

,y r−<br />

i<br />

, 0}, (4.7)<br />

where β r+1<br />

i shows how much is still available <strong>of</strong> bound i in the forward<br />

(positive) direction.<br />

We next increase the counter r by one and repeat (4.4)–(4.7). This takes<br />

care <strong>of</strong> the piecewise linear functions in ξ.<br />

Note that it is possible to solve (4.4) and (4.5) by parametric linear<br />

programming, thereby getting not just one linear piece above E ˜ξ and one<br />

below, but rather piecewise linearity on both sides. Then (4.6) and (4.7) must<br />

be updated to “worst case” analysis <strong>of</strong> bound usage. That is simple to do.<br />

Let us turn to our example (4.1). Since we have developed the piecewise<br />

linear upper bound for equality constraints, we shall repeat the problem with<br />

slack variables added explicitly.<br />

φ(ξ 1 ,ξ 2 )=min{2x raw1 +3x raw2 }<br />

s.t. x raw1 + x raw2 + s 1 = 100,<br />

2x raw1 +6x raw2 − s 2 = 180 + ξ 1 ,<br />

3x raw1 +3x raw2 − s 3 = 162 + ξ 2 ,<br />

x raw1 ≥ 0,<br />

x raw2 ≥ 0,<br />

s 1 ≥ 0,<br />

s 2 ≥ 0,<br />

s 3 ≥ 0.<br />

In this setting, what we need to develop is the following:<br />

{ {<br />

d<br />

+<br />

U(ξ 1 ,ξ 2 )=φ(0, 0) + 1 ξ 1 if ξ 1 ≥ 0, d<br />

d − 1 ξ 1 if ξ 1 < 0, + +<br />

2 ξ 2 if ξ 2 ≥ 0,<br />

d − 2 ξ 2 if ξ 2 < 0.<br />

First, we have already calculated φ(0, 0) = 126 with x raw1 = 36, x raw2 =<br />

18 and s 1 = 46. Next, let us try to find d ± 1 . To do that, we need α1 ,

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!